1; RUN: llc < %s -march=x86-64 -mattr=+mmx,+sse2 | FileCheck %s 2 3define i32 @t0(i64 %x) { 4; CHECK-LABEL: t0: 5; CHECK: # BB#0:{{.*}} %entry 6; CHECK: movd %[[REG1:[a-z]+]], %mm0 7; CHECK-NEXT: pshufw $238, %mm0, %mm0 8; CHECK-NEXT: movd %mm0, %eax 9; CHECK-NEXT: retq 10entry: 11 %0 = bitcast i64 %x to <4 x i16> 12 %1 = bitcast <4 x i16> %0 to x86_mmx 13 %2 = tail call x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx %1, i8 -18) 14 %3 = bitcast x86_mmx %2 to <4 x i16> 15 %4 = bitcast <4 x i16> %3 to <1 x i64> 16 %5 = extractelement <1 x i64> %4, i32 0 17 %6 = bitcast i64 %5 to <2 x i32> 18 %7 = extractelement <2 x i32> %6, i32 0 19 ret i32 %7 20} 21 22define i64 @t1(i64 %x, i32 %n) { 23; CHECK-LABEL: t1: 24; CHECK: # BB#0:{{.*}} %entry 25; CHECK: movd %[[REG2:[a-z]+]], %mm0 26; CHECK-NEXT: movd %[[REG1]], %mm1 27; CHECK-NEXT: psllq %mm0, %mm1 28; CHECK-NEXT: movd %mm1, %rax 29; CHECK-NEXT: retq 30entry: 31 %0 = bitcast i64 %x to x86_mmx 32 %1 = tail call x86_mmx @llvm.x86.mmx.pslli.q(x86_mmx %0, i32 %n) 33 %2 = bitcast x86_mmx %1 to i64 34 ret i64 %2 35} 36 37define i64 @t2(i64 %x, i32 %n, i32 %w) { 38; CHECK-LABEL: t2: 39; CHECK: # BB#0:{{.*}} %entry 40; CHECK: movd %[[REG4:[a-z]+]], %mm0 41; CHECK-NEXT: movd %[[REG6:[a-z0-9]+]], %mm1 42; CHECK-NEXT: psllq %mm0, %mm1 43; CHECK-NEXT: movd %[[REG1]], %mm0 44; CHECK-NEXT: por %mm1, %mm0 45; CHECK-NEXT: movd %mm0, %rax 46; CHECK-NEXT: retq 47entry: 48 %0 = insertelement <2 x i32> undef, i32 %w, i32 0 49 %1 = insertelement <2 x i32> %0, i32 0, i32 1 50 %2 = bitcast <2 x i32> %1 to x86_mmx 51 %3 = tail call x86_mmx @llvm.x86.mmx.pslli.q(x86_mmx %2, i32 %n) 52 %4 = bitcast i64 %x to x86_mmx 53 %5 = tail call x86_mmx @llvm.x86.mmx.por(x86_mmx %4, x86_mmx %3) 54 %6 = bitcast x86_mmx %5 to i64 55 ret i64 %6 56} 57 58define i64 @t3(<1 x i64>* %y, i32* %n) { 59; CHECK-LABEL: t3: 60; CHECK: # BB#0:{{.*}} %entry 61; CHECK: movq (%[[REG1]]), %mm0 62; CHECK-NEXT: psllq (%[[REG3:[a-z]+]]), %mm0 63; CHECK-NEXT: movd %mm0, %rax 64; CHECK-NEXT: retq 65entry: 66 %0 = bitcast <1 x i64>* %y to x86_mmx* 67 %1 = load x86_mmx, x86_mmx* %0, align 8 68 %2 = load i32, i32* %n, align 4 69 %3 = tail call x86_mmx @llvm.x86.mmx.pslli.q(x86_mmx %1, i32 %2) 70 %4 = bitcast x86_mmx %3 to i64 71 ret i64 %4 72} 73 74declare x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx, i8) 75declare x86_mmx @llvm.x86.mmx.pslli.q(x86_mmx, i32) 76declare x86_mmx @llvm.x86.mmx.por(x86_mmx, x86_mmx) 77 78